Practical Deep Learning Examples with matlab

Common Semantic Segmentation Applications
• Autonomous driving: for identifying a drivable path for cars by separating the road 
from obstacles like pedestrians, sidewalk, poles, and other cars
• Industrial inspection: for detecting defects in materials, such as wafer inspection
• Satellite imagery: for identifying mountains, rivers, deserts, and other terrain 
• Medical imaging: for analyzing and detecting cancerous anomalies in cells 

---

17 | Practical Deep Learning Examples with MATLAB
Semantic Segmentation Network Architecture
As we saw in examples 1 and 2, a traditional CNN takes an image, 
passes it through the layers of the network, and then outputs a final 
class.
A semantic segmentation network builds on this process with an
up-sampling network, which has an architecture similar to a
reversed CNN.
This series of new layers 
upsamples 
the result of the pretrained network 
back into the image. The result is an image with every pixel assigned a 
classification label.

---

18 | Practical Deep Learning Examples with MATLAB
1. Importing a Pretrained Network
In this example we want to build a network that an autonomous driving 
system can use to detect clear road space, driving lanes, and sidewalks. 
The steps are as follows: 
1. Import a pretrained network.
2. Load in the dataset.
3. Set up the network.
4. Train the network.
5. Evaluate the network’s accuracy.
We could train a network from scratch, but for this example, we’ll use 
a pretrained network—as we saw in the previous example, we can test 
a pretrained network on images in the categories that it was originally 
trained on without reconfiguring it. 
Our pretrained network is the VGG-16. Used in the 
ImageNet
Large-Scale Visual Recognition Challenge
 (ILSVRC), VGG-16 is trained 
on more than a million images and can classify images into 1000 ob-
ject categories. 
Importing the VGG-16 takes just one line of MATLAB code: